AI Paper Atlas

Research map / 3D vision / NeRF and radiance fields

GRF: Learning a General Radiance Field for 3D Representation and Rendering

Alex Trevithick, Bo Yang

ICCV 2021 · 3D vision › NeRF and radiance fields · approach: Inverse rendering with neural fields · 264 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add input viewpoints to the CNN module (ablation)GRF · ShapeNetv2 Cars (ablation subset)PSNR: 20.13 -> 27.16Improved
add input viewpoints to the CNN module (ablation)GRF · ShapeNetv2 Cars (ablation subset)SSIM: 0.807 -> 0.942Improved
add pixel local features (CNN decoder) instead of a global feature vector (ablation)GRF · ShapeNetv2 Cars (ablation subset)PSNR: 20.23 -> 27.16Improved
add pixel local features instead of a global feature vector (ablation)GRF · ShapeNetv2 Cars (ablation subset)SSIM: 0.818 -> 0.942Improved
use attention aggregation instead of max-pooling (ablation)GRF · ShapeNetv2 Cars (ablation subset)PSNR: 24.88 -> 27.16Improved
use attention aggregation instead of max-pooling (ablation)GRF · ShapeNetv2 Cars (ablation subset)SSIM: 0.914 -> 0.942Improved
increase input views from 2 to 6GRF · ShapeNet unseen 7 classesPSNR: 24.68 -> 29.37Improved
increase input views from 2 to 6 for SRNs (retrained)SRNs · ShapeNet unseen 7 classesPSNR: 24.16 -> 25.76Improved
finetune pretrained GRF for 100 iterations instead of training NeRF from scratchGRF · Synthetic-NeRF (4 novel scenes)PSNR: 15.15 -> 19.69Improved
finetune pretrained GRF for 10k iterations instead of training NeRF from scratchGRF · Synthetic-NeRF (4 novel scenes)PSNR: 23.35 -> 25.10Improved
finetune pretrained GRF for 1k iterations vs noneGRF · Synthetic-NeRF (4 novel scenes)LPIPS: 0.246 -> 0.128Improved
use GRF instead of NeRF for single-scene learningGRF vs NeRF · Real-world scenes (8 scenes)LPIPS: 0.250 -> 0.178Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy